Physics-Informed Neural Network-Based Persistent Optimal Control of Sustainably Powered Mobile Robots

Explore the Learned Value Function

This interactive visualization is an early look at our ongoing work on physics-informed neural networks for persistent robot control. It shows the network’s predicted discounted value as a function of robot position and battery state, for different information states.

Rotate the surface by dragging, hover to inspect values, and choose a clarity preset. To explore a custom information state, adjust the clarity values at the points of interest and select Update surface. These are predictions from a research model in development.

Open interactive visualization in a new tab ↗

For a larger workspace, use the new-tab link above. Custom clarity values are evaluated locally in your browser.